基于BP神经网络机器人实时避障算法

针对二维静态环境下智能机器人避障及路径规划问题,提出了基于BP神经网络的机器人实时避障算法。首先,用多个扇区表示机器人周围的环境,利用激光雷达探测每个扇区内障碍物的距离信息,以每个扇区内障碍物的距离信息为输入,利用BP神经网络计算该扇区被选择为避障方向的得分;然后,利用各扇区中点坐标与当前时刻距障碍物最近扇区中点坐标之间的欧氏距离,计算机器人在当前位姿条件下各扇区被选中作为避障方向的条件概率;最后,将使得得分与条件概率之积最大的扇区作为机器人的避障方向。实验结果表明:所提算法的收敛时间比栅格方法降低了50%以上,机器人的避障轨迹与人工势场方法相比更短,能较好地应用于复杂多障碍物场景。

Real time obstacle avoidance algorithm for robots based on BP neural network

To address the problem of obstacle avoidance and path planning of intelligent robots in two dimensional static environment, a real time obstacle avoidance algorithm based on BP neural network is proposed. Firstly, multiple sectors are used to represent the environment around the robot, and lidar is utilized to detect the distance information of obstacles in each sector. With the distance information of obstacles in each sector, BP neural network is used to calculate the score of the sector selected as obstacle avoidance direction. Then, the Euclidean distance between the mid point coordinate of each sector and the mid point coordinate of the closest sector to the obstacle at the current moment is used to calculate the conditional probability. Each sector is selected as the direction of obstacle avoidance under the current pose of the robot. Finally, the sector with the largest product of score and conditional probability is taken as the obstacle avoidance direction of the robot. Experimental results show that the convergence time of the proposed algorithm is 50% less than that of the grid method, and the obstacle avoidance trajectory of the robot is shorter than that of the artificial potential field method. It can be better applied to complex multi obstacle scenarios.